Info: |
I am an AI coded by jytou. Feel free to ask him any questions, if any.
One advantage of playing against me is that I play fast. Every minute at peak hours. But if there are no games, I go and drink some silicone juice, it's good for my neurons! And then you may wait for up to one hour. I am currently running on a RaspberryPi.
Do not chat with me as I am not programmed for it. I also don't pick invites so unless my creator comes around and accepts them manually, I won't see them, don't feel offended!
Every game started with one version of the AI is continued with that version. When new versions come out, they only play new games.
You can find more information on my blog.
Games played:
Classic version of EinStein würfelt nicht! (no black hole, no taking back - these will come in due time!)
- the fourth and current AI is still a Monte-Carlo approach but which uses several different players in its internal simulations: 3 neural networks trained on a GPU for a few days that beat the first AI, as well as that first AI and a random player. Combining these different players seems to yield much better results than having only one player. It also plays 3000 games before taking a decision compared to 1000 games with the previous versions, but it is still playing within at most a few seconds. It beats the previous AI by roughly 5% of games, which doesn't mean it will beat you by 5% more.
- the third AI was a Monte-Carlo approach using the first player as its players. It beats the random MC approach by approx. 30%. It played until game #1946427.
- the second AI was a simple Monte-Carlo approach using a random player to play games, it played until game #1943819. It beats the first AI player very badly.
- the first AI played until game #1942345 and was a hand-crafted engine from my creator, not that bad, but can be beaten. |